feat: wire vLLM SageMaker entrypoint to standard-supervisor#6044
Merged
Conversation
Update sagemaker_entrypoint.sh to use standard-supervisor for process management and automatic requirements.txt installation. Changes: - sagemaker_entrypoint.sh: exec standard-supervisor instead of raw exec - Dockerfile: bump model-hosting-container-standards pin to >=0.1.15 Signed-off-by: Loki Ravi <lokravi@amazon.com>
The entrypoint now uses 'exec standard-supervisor python3 ...' instead of 'exec python3 ...'. Update the dry-run regex in TestEntrypointArgHandling to match both formats. Signed-off-by: Loki Ravi <lokravi@amazon.com>
log4j-core 2.17.1 is bundled in ray_dist.jar and cannot be patched without a Ray upgrade. Extend review_by to 2026-06-04. Signed-off-by: Loki Ravi <lokravi@amazon.com>
rustls-webpki 0.103.12 is statically linked in /usr/local/bin/uv. Fix requires rustls-webpki>=0.104.0-alpha.7 (pre-release, not yet available in stable uv). Not exploitable in our usage context. Signed-off-by: Loki Ravi <lokravi@amazon.com>
junpuf
reviewed
May 6, 2026
…ckerfile Align the AL2023 vLLM image with the Ubuntu image by requiring model-hosting-container-standards>=0.1.15 for standard-supervisor dependency auto-install support. Signed-off-by: Loki Ravi <lokravi@amazon.com>
Same rustls-webpki vulnerability in uv binary, same justification as the vllm (Ubuntu) allowlist entry. Signed-off-by: Loki Ravi <lokravi@amazon.com>
junpuf
approved these changes
May 8, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Adds standard-supervisor (from model-hosting-container-standards) to the vLLM SageMaker launch chain, aligning the DLC with the vLLM open source SageMaker stage which already uses standard-supervisor as its entrypoint wrapper.
Why this is necessary
The upstream vLLM project's sagemaker-entrypoint.sh already launches via standard-supervisor.
Without standard-supervisor, if the vLLM process crashes, the container exits immediately. With it, the process is managed by supervisord which:
Automatically restarts the vLLM process on unexpected exits (configurable via PROCESS_AUTO_RECOVERY, default: true)
Retries up to N times before giving up (configurable via PROCESS_MAX_START_RETRIES, default: 3)
Provides structured exit codes so SageMaker can distinguish between transient failures and permanent ones
standard-supervisor (v0.1.15+) automatically installs dependencies from requirements.txt before starting the server. This enables customers to bundle custom dependencies with their model artifacts without needing a custom container.
Controlled by:
STANDARD_AUTO_INSTALL_REQ (default: true) — enable/disable
STANDARD_PIP_ARGS — override with explicit pip arguments
Testing
Local validation
CI will run: sanity tests, security tests, telemetry tests, upstream vLLM tests (GPU), and a real SageMaker endpoint deployment test